Intelligent diagnosis system, method, medium and equipment

Abstract

The invention relates to an intelligent diagnosis system and method, a medium and equipment, and the system comprises a data collection module which collects a multi-mode signal; the feature extraction module is used for extracting features of each modal signal from the preprocessed multi-modal signals; the feature extraction module is configured with a plurality of mutually independent 1D-CNNs, each modal signal is respectively input into the corresponding 1D-CNNs, and features of each modal signal are extracted; the gating fusion module is used for dynamically weighting and integrating each modal signal to obtain a multi-modal fusion feature based on a gating fusion mechanism and the correlation between each modal signal; the double-attention module is used for dynamically adjusting attention distribution in the multi-modal fusion features based on a channel-time double-domain attention mechanism and pathological complexity information, so that the recognition capability of complex cases is improved; and the diagnosis prediction module is used for outputting a multi-label disease prediction value by taking the multi-modal fusion features after the attention distribution is dynamically adjusted as input.

Description

Technical Field

[0001] This disclosure relates to the field of intelligent healthcare, specifically to an intelligent diagnostic system, method, medium, and device. Background Technology

[0002] Cardiovascular disease (CVD) is one of the leading causes of death among chronic diseases worldwide, and its early and accurate diagnosis is crucial for reducing mortality and improving patient prognosis. Traditional diagnosis of cardiovascular disease relies on physicians' empirical interpretation of single-modality physiological signals (such as electrocardiograms), which is easily affected by subjective judgment, signal noise, and individual differences, leading to the risk of missed diagnoses and misdiagnoses.

[0003] In recent years, artificial intelligence technology has demonstrated great potential in medical signal processing and analysis, particularly in the automatic identification of arrhythmias, heart failure, coronary heart disease, and valvular diseases. However, current technologies still have the following shortcomings: 1) Single multimodal signal fusion method: Existing methods usually use early feature splicing or late decision fusion, which fails to fully explore the spatiotemporal correlation and complementary information between different modalities, and is prone to information redundancy or loss of key signals; 2) The attention mechanism lacks dynamic adaptability: Existing attention modules such as SE and CBAM are statically designed and cannot adaptively adjust the attention weight according to the pathological characteristics of the input signal, making it difficult to handle individual differences and complex pathologies. 3) Signal quality and pathological significance are not jointly modeled: Clinical physiological signals are often affected by motion artifacts, baseline drift, sensor noise, etc., and existing models have failed to effectively distinguish noise from real pathological features; 4) Limited diagnostic applicability: Most models only identify specific diseases (such as atrial fibrillation) and lack the ability to identify a wide range of cardiovascular diseases.

[0004] Therefore, there is an urgent need for technical solutions that can overcome the above-mentioned shortcomings. Summary of the Invention

[0005] To address the problems existing in the prior art, this disclosure proposes an intelligent diagnostic system, method, medium, and device to solve at least one of the technical problems listed in the background art. The technical solution adopted in this disclosure is as follows: In a first aspect, this disclosure provides an intelligent diagnostic system, the system comprising: The data acquisition module is used to synchronously acquire multimodal signals related to the patient's health, including modal signals such as ECG, PPG, PCG, and BP. The feature extraction module is used to extract local and global features of each modality signal from the preprocessed multimodal signal. The feature extraction module is configured with several independent one-dimensional convolutional neural networks (1D-CNNs). Each modality signal is input into a corresponding 1D-CNN and the corresponding 1D-CNN extracts local and global features. The gated fusion module is used to dynamically weight and integrate the local and global features of each modal signal based on the gated fusion mechanism and the correlation between each modal signal to obtain multimodal fusion features; A dual attention module is used to dynamically adjust the attention distribution in the multimodal fusion features based on a channel-time dual-domain attention mechanism and pathological complexity information, thereby improving the ability to identify complex cases. The diagnostic prediction module takes the multimodal fusion features after dynamically adjusting the attention distribution as input and outputs multi-label disease prediction values.

[0006] Preferably, the system further includes: The data preprocessing module is used to preprocess the multimodal signal, including noise reduction, signal alignment and normalization.

[0007] Preferably, each modal signal is input to a corresponding 1D-CNN and the corresponding 1D-CNN extracts local and global features. The expression can be as follows: ; in, : Represents a specific modal signal, that is, the i-th modal signal from sampling; i : Represents the number of the modal signal; CNN i : Represents the first i 1D-CNN corresponding to each modal signal; , representing the i Feature maps extracted from each modal signal For time step, For feature dimensions.

[0008] In practice, a typical CNN structure consists of 3 convolutional layers (kernel=3, stride=1), each followed by BatchNorm and ReLU activation. The convolutional kernels can be expanded or deep networks can be stacked to extract multi-scale features, depending on the actual signal length and sampling rate.

[0009] Preferably, the gating fusion weights of the gating fusion mechanism are dynamically generated by the correlation between the modal signals, and the expression can be as follows: ; in, , representing the gating fusion weight, For time step, For feature dimensions; The weight matrix is ​​a learnable weight matrix; This indicates splicing along the channel dimension; For the Sigmoid function; bg It is the bias vector; F ECG F PPG F PCG and F BP The extracted feature maps are ECG, PPG, PCG and BP, respectively. The extracted feature maps include local features and global features.

[0010] Preferably, the expression for the multimodal fusion feature can be as follows: ; in, This indicates element-wise multiplication; i The number representing the modal signal; F i Representing the i Feature maps extracted from each modal signal; G i Representing the i Gated fusion weights corresponding to each modal signal; F fused This represents the obtained multimodal fusion features.

[0011] In practice, the gated fusion mechanism can automatically reduce the weight of low-quality modal signals (such as PPG with severe motion artifacts), enhance the contribution of high-confidence modal signals, and improve diagnostic robustness in complex environments.

[0012] Preferably, the channel-time dual-domain attention mechanism includes channel attention and time attention.

[0013] Preferably, the expression for the channel attention is: ; The expression for the time attention is: ; in, For channel attention, For time attention; , All of these are learnable parameters. This refers to the channel compression ratio. For feature dimension, For time step, Represents the number of hidden states; F is the temperature coefficient. fused The multimodal fusion feature; Based on the base temperature, To control the intensity of pathological regulation of attention.

[0014] By incorporating channel attention and temporal attention, this disclosure introduces pathological complexity into the attention mechanism for the first time, enabling adaptive extraction of dynamic features from both simple and complex cases.

[0015] Furthermore, channel attention A lightweight pathology complexity classifier can be configured. Furthermore, they are classified according to the complexity of the pathology.

[0016] Preferably, the expression for the temperature coefficient is: ; in, For temperature coefficient, Based on the base temperature, Controlling the intensity of pathology's regulation of attention; To predict complexity, .

[0017] By temperature coefficient This can avoid local overfitting caused by increased pathological complexity; for example, when pathological complexity increases, the temperature coefficient can be increased accordingly. This makes the attention distribution of the channel-time dual-domain attention mechanism smoother and avoids local overfitting.

[0018] Preferably, the diagnostic prediction module 600 is configured with a preset diagnostic model for multi-label disease prediction, the preset diagnostic model including: A Long Short-Term Memory (LSTM) network is used to capture the time dependence of the multimodal fusion features after dynamically adjusting the attention distribution; Fully connected layer, used to output multi-label disease prediction values.

[0019] This disclosure supports broad-spectrum joint screening of cardiovascular diseases through a diagnostic prediction module 600 and a configured preset diagnostic model, with multi-task output. It can simultaneously identify multiple diseases and distinguish complex pathological features, thereby achieving joint diagnosis of multiple cardiovascular diseases such as coronary heart disease, heart failure, arrhythmia, cardiomyopathy, and valvular heart disease.

[0020] A second aspect of this disclosure provides a method for the intelligent diagnostic system provided in the first aspect, the method comprising: The data acquisition module synchronously acquires multimodal signals related to the patient's health, including modal signals such as ECG, PPG, PCG, and BP. The feature extraction module extracts local and global features of each modality signal from the preprocessed multimodal signal. The feature extraction module 300 is configured with several independent one-dimensional convolutional neural networks (1D-CNNs). Each modality signal is input into a corresponding 1D-CNN and the corresponding 1D-CNN extracts local and global features. The gated fusion module dynamically weights and integrates the local and global features of each modal signal based on the gated fusion mechanism and the correlation between each modal signal to obtain multimodal fusion features. The dual attention module dynamically adjusts the attention distribution in the multimodal fusion features based on the channel-time dual-domain attention mechanism and pathological complexity information, thereby improving the ability to identify complex cases. The diagnostic prediction module takes the multimodal fusion features after dynamically adjusting the attention distribution as input and outputs multi-label disease prediction values.

[0021] Preferably, before extracting the local and global features of each modal signal from the preprocessed multimodal signal through the feature extraction module, the method further includes: The multimodal signal is preprocessed, including noise reduction, signal alignment, and normalization.

[0022] In a third aspect, this disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the intelligent diagnostic system and / or a method for the intelligent diagnostic system as described above.

[0023] In a fourth aspect, this disclosure provides an electronic device including a processor and a memory, the processor being configured to execute a computer program stored in the memory to implement the intelligent diagnostic system as described above, or to implement the method as described above.

[0024] The electronic device can be a wearable device, a bedside monitoring system, or a telemedicine platform, which makes the system deployment and clinical application easier, and enables real-time, intelligent cardiovascular health monitoring and diagnosis.

[0025] The beneficial effects of this disclosure are as follows: This disclosure proposes an intelligent diagnostic system, method, medium, and device. By collecting multimodal physiological signals such as electrocardiogram (ECG), pulse wave (PPG), phonocardiogram (PCG), and blood pressure (BP), and combining an adaptive attention mechanism with a multi-task deep neural network model, this disclosure achieves high-precision, low-false-alarm-rate automatic detection and diagnosis of cardiovascular diseases. It is applicable to various application scenarios such as clinical monitoring, wearable devices, and telemedicine.

[0026] In this disclosure, a gating fusion mechanism is used to dynamically weight and integrate the local and global features of each modality based on the correlation between each modality. This allows for dynamic weighting to suppress low-quality modal signals and improve the robustness of multimodal feature fusion.

[0027] In this disclosure, the attention distribution of the channel-time dual-domain attention mechanism is dynamically adjusted based on the multimodal fusion features and pathological complexity information. By adjusting the attention distribution using pathological complexity, the ability to identify complex cases is improved, and optimal feature extraction is achieved for both complex and simple cases.

[0028] In this disclosure, a channel-time dual-domain attention mechanism, a long short-term memory network (LSTM), and a fully connected layer are used to successfully construct an end-to-end multi-task diagnostic architecture, which realizes integrated signal preprocessing, feature extraction, dynamic fusion, and multi-task classification, and is suitable for wearable devices and clinical monitoring systems.

[0029] This disclosure has broad applications in: (A) smart wearable devices for real-time monitoring of heart rhythm and function, providing early disease warnings; (B) emergency departments and intensive care units, assisting in the rapid identification of acute coronary syndromes and severe arrhythmias; (C) cardiac rehabilitation and chronic disease management, providing long-term dynamic monitoring of changes in patient cardiac function and optimizing treatment plans; and (D) telemedicine platforms, providing high-precision auxiliary diagnostic tools for primary healthcare institutions and promoting the popularization of smart healthcare. This disclosure significantly improves the intelligence level of multimodal cardiovascular signal analysis, possessing high clinical translational value and broad market prospects. Attached Figure Description

[0030] The accompanying drawings, which form part of this application, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0031] Figure 1This is an architecture diagram of an intelligent diagnostic system provided in Embodiment 1 of this disclosure; wherein, the data preprocessing module is an optional module.

[0032] Figure 2 This is a flowchart of a method for the intelligent diagnostic system described in Embodiment 1, provided in Embodiment 2 of this disclosure; wherein, S200 is an optional step. Detailed Implementation

[0033] The present disclosure will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0034] The following detailed descriptions are exemplary and intended to provide further detailed explanation of this disclosure. Unless otherwise specified, all technical terms used in this disclosure have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure.

[0035] Terminology Explanation: ECG: Electrocardiogram, electrocardiogram.

[0036] PPG: Photo-Plethysmography, also known as "pulse wave".

[0037] PCG: phonocardiogram.

[0038] BP: Blood Pressure.

[0039] Example 1: like Figure 1 As shown, this disclosure provides an intelligent diagnostic system, the system comprising: The data acquisition module 100 is used to synchronously acquire multimodal signals related to the patient's health, including modal signals such as ECG, PPG, PCG and BP. The feature extraction module 300 is used to extract local and global features of each modality signal from the preprocessed multimodal signal. The feature extraction module 300 is configured with several independent one-dimensional convolutional neural networks (1D-CNNs). Each modality signal is input into a corresponding 1D-CNN and the corresponding 1D-CNN extracts local and global features. The gated fusion module 400 is used to dynamically weight and integrate the local and global features of each modal signal based on the gated fusion mechanism and the correlation between each modal signal to obtain multimodal fusion features; The dual attention module 500 is used to dynamically adjust the attention distribution in the multimodal fusion features based on the channel-time dual-domain attention mechanism and pathological complexity information, thereby improving the ability to identify complex cases. The diagnostic prediction module 600 is used to take the multimodal fusion features after dynamically adjusting the attention distribution as input and output multi-label disease prediction values.

[0040] Furthermore, the ECG acquisition method can be: 12 leads are used for acquisition, with a sampling rate of 500 Hz.

[0041] Furthermore, the PPG acquisition method can be: using a fingertip or wrist photoplethysmography sensor for acquisition, with a sampling rate of 100 Hz.

[0042] Furthermore, the PCG acquisition method can be: using an electronic stethoscope for acquisition, with a sampling rate of 2000 Hz; Furthermore, BP can be acquired using oscillometric or intra-arterial pressure monitoring methods at a sampling rate of 1 Hz (continuous or intermittent sampling is possible).

[0043] In one feasible implementation, the system further includes: The data preprocessing module 200 is used to preprocess the multimodal signal, including noise reduction, signal alignment and standardization.

[0044] Furthermore, the denoising includes using wavelet transform (db6) to remove ECG baseline drift and electromyographic interference.

[0045] Furthermore, the signal alignment includes resampling each modal signal to a unified time axis (500 Hz) using linear interpolation with the R-wave peak value of the ECG as the time reference.

[0046] Furthermore, the standardization process employs the Z-score normalization formula as follows: ; in: For the first Mode in time The original signal value; , These are the mean and standard deviation of the signal segment for that modality, respectively. This is the normalized signal.

[0047] In one feasible implementation, each modal signal is input to a corresponding 1D-CNN and the corresponding 1D-CNN extracts local and global features, as shown in the following expression: ; in, : Represents a specific modal signal, that is, the i-th modal signal from sampling; i : Represents the number of the modal signal; CNN i : Represents the first i 1D-CNN corresponding to each modal signal; , representing the i Feature maps extracted from each modal signal For time step, For feature dimensions.

[0048] In practice, a typical CNN structure consists of 3 convolutional layers (kernel=3, stride=1), each followed by BatchNorm and ReLU activation. The convolutional kernels can be expanded or deep networks can be stacked to extract multi-scale features, depending on the actual signal length and sampling rate.

[0049] In one feasible implementation, the gating fusion weights of the gating fusion mechanism are dynamically generated by the correlation between the modal signals, and the expression can be as follows: ; in, , representing the gating fusion weight, For time step, For feature dimensions; The weight matrix is ​​a learnable weight matrix; This indicates splicing along the channel dimension; For the Sigmoid function; bg It is the bias vector; F ECG F PPG F PCG and F BP The extracted feature maps are ECG, PPG, PCG and BP, respectively. The extracted feature maps include local features and global features.

[0050] Furthermore, the expression for the multimodal fusion feature can be as follows: ; in, This indicates element-wise multiplication; i The number representing the modal signal; F i Representing thei Feature maps extracted from each modal signal; G i Representing the i Gated fusion weights corresponding to each modal signal; F fused This represents the obtained multimodal fusion features.

[0051] In practice, the gated fusion mechanism can automatically reduce the weight of low-quality modal signals (such as PPG with severe motion artifacts), enhance the contribution of high-confidence modal signals, and improve diagnostic robustness in complex environments.

[0052] In one feasible implementation, the channel-temporal dual-domain attention mechanism includes channel attention and temporal attention.

[0053] Furthermore, the expression for the channel attention is: ; The expression for the time attention is: ; in, For channel attention, For time attention; , All of these are learnable parameters. This refers to the channel compression ratio. For feature dimension, For time step, This represents the number of hidden states (this is common knowledge regarding attention mechanisms). F is the temperature coefficient. fused The multimodal fusion feature; Based on the base temperature, To control the intensity of pathological regulation of attention.

[0054] By incorporating channel attention and temporal attention, this disclosure introduces pathological complexity into the attention mechanism for the first time, enabling adaptive extraction of dynamic features from both simple and complex cases.

[0055] Furthermore, channel attention A lightweight pathology complexity classifier can be configured. Furthermore, they are classified according to the complexity of the pathology.

[0056] Furthermore, the expression for the temperature coefficient is: ; in, For temperature coefficient, Based on the base temperature, Controlling the intensity of pathology's regulation of attention; To predict complexity, .

[0057] By temperature coefficient This can avoid local overfitting caused by increased pathological complexity; for example, when pathological complexity increases, the temperature coefficient can be increased accordingly. This makes the attention distribution of the channel-time dual-domain attention mechanism smoother and avoids local overfitting.

[0058] In one feasible implementation, the diagnostic prediction module 600 is configured with a preset diagnostic model for multi-label disease prediction, the preset diagnostic model including: A Long Short-Term Memory (LSTM) network is used to capture the time dependence of the multimodal fusion features after dynamically adjusting the attention distribution; Fully connected layer, used to output multi-label disease prediction values.

[0059] Furthermore, the expression for the multi-label disease prediction value is as follows: ; in, This refers to the hidden state of the Long Short-Term Memory (LSTM) network at the last time step. , These are the output weights and biases, respectively. Number of disease categories (Example: 6 categories, diseases can include normal, atrial fibrillation, heart failure, myocardial ischemia, ventricular tachycardia, valvular disease); This is a multi-label disease predictor.

[0060] This disclosure supports broad-spectrum joint screening of cardiovascular diseases through a diagnostic prediction module 600 and a configured preset diagnostic model, with multi-task output. It can simultaneously identify multiple diseases and distinguish complex pathological features, thereby achieving joint diagnosis of multiple cardiovascular diseases such as coronary heart disease, heart failure, arrhythmia, cardiomyopathy, and valvular heart disease.

[0061] The diagnostic prediction module 600 can be used to perform multi-task classification on multimodal fusion features, thereby obtaining multi-label disease prediction values. The multi-task output supports broad-spectrum joint screening of cardiovascular diseases, and can simultaneously identify multiple diseases and distinguish complex pathological features.

[0062] Example 2: like Figure 1 As shown, this disclosure provides a method for the intelligent diagnostic system described in Embodiment 1, the method comprising: S100. The data acquisition module 100 synchronously acquires multimodal signals related to the patient's health, including modal signals such as ECG, PPG, PCG and BP. S300, The feature extraction module 300 extracts local and global features of each modality signal from the preprocessed multimodal signal; the feature extraction module 300 is configured with several independent one-dimensional convolutional neural networks (1D-CNNs), and each modality signal is input into a corresponding 1D-CNN and the corresponding 1D-CNN extracts local and global features; S400: Based on the gating fusion mechanism and the correlation between each modal signal, the gating fusion module 400 dynamically weights and integrates the local and global features of each modal signal to obtain multimodal fusion features. S500: The dual attention module 500 dynamically adjusts the attention distribution in the multimodal fusion features based on the channel-time dual-domain attention mechanism and pathological complexity information, thereby improving the ability to identify complex cases. S600: The diagnostic prediction module 600 takes the multimodal fusion features after dynamically adjusting the attention distribution as input and outputs a multi-label disease prediction value.

[0063] Preferably, before the feature extraction module 300 extracts the local and global features of each modal signal from the preprocessed multimodal signal, the method further includes: S200. Preprocess the multimodal signal, including noise reduction, signal alignment and normalization.

[0064] S100, S200, S300, S400, S500 and S600 correspond to the data acquisition module 100, the data preprocessing module 200, the feature extraction module 300, the gating fusion module 400, the dual attention module 500 and the diagnostic prediction module 600, respectively.

[0065] Example 3: Embodiment 3 of this disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in Embodiment 2. Alternatively, an intelligent diagnostic system as described in Example 1 can be implemented.

[0066] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules, or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory or other memory technologies, CD-ROM (Compact Disc Read-Only Memory), DVD or other optical disc storage, cartridges, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer.

[0067] Example 4: Embodiment 4 of this disclosure provides an electronic device including a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the intelligent diagnostic system described in Embodiment 1, or to implement the method for the intelligent diagnostic system described in Embodiment 1 as described in Embodiment 2.

[0068] The electronic device can be a wearable device, a bedside monitoring system, or a telemedicine platform, which makes the system deployment and clinical application easier, and enables real-time, intelligent cardiovascular health monitoring and diagnosis.

[0069] Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0070] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, approaches, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0071] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (methods), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0073] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0075] In summary, the intelligent diagnostic system, method, medium, and device provided in embodiments 1-4 of this disclosure achieve high-precision, low-false-report automatic detection and diagnosis of cardiovascular diseases by collecting multimodal physiological signals such as electrocardiogram (ECG), pulse wave (PPG), phonocardiogram (PCG), and blood pressure (BP), and combining an adaptive attention mechanism with a multi-task deep neural network model. This makes it suitable for various application scenarios such as clinical monitoring, wearable devices, and telemedicine. This disclosure can be widely applied to: (A) intelligent wearable devices for real-time monitoring of heart rhythm and function, providing early disease warnings; (B) emergency departments and intensive care units to assist in the rapid identification of acute coronary syndromes and severe arrhythmias; (C) cardiac rehabilitation and chronic disease management to monitor changes in patient cardiac function over a long period and optimize treatment plans; and (D) telemedicine platforms to provide high-precision auxiliary diagnostic tools for primary healthcare institutions, promoting the popularization of intelligent healthcare. This disclosure significantly improves the intelligence level of multimodal cardiovascular signal analysis, possessing high clinical translational value and broad market prospects.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure and not to limit them. Although this disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of this disclosure. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this disclosure should be covered within the protection scope of the claims of this disclosure.

Claims

1. An intelligent diagnostic system, characterized in that, The system includes: The data acquisition module is used to synchronously acquire multimodal signals related to the patient's health, including modal signals such as ECG, PPG, PCG and BP. The feature extraction module is used to extract local and global features of each modality signal from the preprocessed multimodal signal. The feature extraction module is configured with several independent one-dimensional convolutional neural networks (1D-CNNs). Each modality signal is input into a corresponding 1D-CNN and the corresponding 1D-CNN extracts local and global features. The gated fusion module is used to dynamically weight and integrate the local and global features of each modal signal based on the gated fusion mechanism and the correlation between each modal signal to obtain multimodal fusion features; A dual-attention module is used to dynamically adjust the attention distribution in the multimodal fusion features based on a channel-time dual-domain attention mechanism and pathological complexity information; The diagnostic prediction module takes the multimodal fusion features after dynamically adjusting the attention distribution as input and outputs multi-label disease prediction values.

2. The intelligent diagnostic system as described in claim 1, characterized in that, The system also includes: The data preprocessing module is used to preprocess the multimodal signal, including noise reduction, signal alignment and normalization.

3. The intelligent diagnostic system as described in claim 1, characterized in that, Each modal signal is input into a corresponding 1D-CNN and the corresponding 1D-CNN extracts local and global features, as shown in the following expressions: ； in, : Represents a specific modal signal, that is, the i-th modal signal from sampling; i : Represents the number of the modal signal; CNN i : Represents the first i 1D-CNN corresponding to each modal signal; , representing the i Feature maps extracted from each modal signal For time step, For feature dimensions.

4. The intelligent diagnostic system as described in claim 1, characterized in that, The gate fusion weights of the gated fusion mechanism are dynamically generated based on the correlation between the signals of each modality, as expressed below: ； in, , representing the gating fusion weight, For time step, For feature dimensions; The weight matrix is ​​a learnable weight matrix; This indicates splicing along the channel dimension; For the Sigmoid function; bg It is the bias vector; F ECG F PPG F PCG and F BP The extracted feature maps are ECG, PPG, PCG and BP, respectively. The extracted feature maps include local features and global features. The expression for the multimodal fusion feature is as follows: ； in, This indicates element-wise multiplication; i The number representing the modal signal; F i Representing the i Feature maps extracted from each modal signal; G i Representing the i Gated fusion weights corresponding to each modal signal; F fused This represents the obtained multimodal fusion features.

5. The intelligent diagnostic system as described in claim 1, characterized in that, The channel-temporal dual-domain attention mechanism includes channel attention and temporal attention; The expression for the channel attention is: ; The expression for the time attention is: ; in, For channel attention, For time attention; , All of these are learnable parameters. This refers to the channel compression ratio. For feature dimension, For time step, Represents the number of hidden states; F is the temperature coefficient. fused The multimodal fusion feature; Based on the base temperature, To control the intensity of pathological regulation of attention.

6. The intelligent diagnostic system as described in claim 1, characterized in that, The expression for the temperature coefficient is: ; in, For temperature coefficient, Based on the base temperature, Controlling the intensity of pathology's regulation of attention; To predict complexity, .

7. The intelligent diagnostic system as described in claim 1, characterized in that, The diagnostic prediction module 600 is configured with a preset diagnostic model for multi-label disease prediction, the preset diagnostic model including: A Long Short-Term Memory (LSTM) network is used to capture the time dependence of the multimodal fusion features after dynamically adjusting the attention distribution; Fully connected layer, used to output multi-label disease prediction values; The expression for the multi-label disease prediction value is as follows: ； in, This refers to the hidden state of the Long Short-Term Memory (LSTM) network at the last time step. , These are the output weights and biases, respectively. Number of disease categories; This is a multi-label disease predictor.

8. A method for use in the intelligent diagnostic system as described in any one of claims 1-7, characterized in that, The method includes: The data acquisition module synchronously acquires multimodal signals related to the patient's health, including modal signals such as ECG, PPG, PCG, and BP. The feature extraction module extracts local and global features of each modality signal from the preprocessed multimodal signal. The feature extraction module 300 is configured with several independent one-dimensional convolutional neural networks (1D-CNNs). Each modality signal is input into a corresponding 1D-CNN and the corresponding 1D-CNN extracts local and global features. The gated fusion module dynamically weights and integrates the local and global features of each modal signal based on the gated fusion mechanism and the correlation between each modal signal to obtain multimodal fusion features. The dual attention module dynamically adjusts the attention distribution in the multimodal fusion features based on the channel-time dual-domain attention mechanism and pathological complexity information, thereby improving the ability to identify complex cases. The diagnostic prediction module takes the multimodal fusion features after dynamically adjusting the attention distribution as input and outputs multi-label disease prediction values.

9. The method as described in claim 8, characterized in that, Before the feature extraction module 300 extracts the local and global features of each modal signal from the preprocessed multimodal signal, the method further includes: The multimodal signal is preprocessed, including noise reduction, signal alignment, and normalization.

10. An electronic device comprising a processor and a memory, characterized in that, The processor is used to execute computer programs stored in the memory to implement the intelligent diagnostic system as described in any one of claims 1-7, or to implement the method as described in any one of claims 8-9.

Images